1
0
Fork 1
mirror of https://github.com/NixOS/nixpkgs.git synced 2024-11-18 19:51:17 +00:00
nixpkgs/nixos/tests/acme.nix

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

717 lines
28 KiB
Nix
Raw Normal View History

{ pkgs, lib, ... }: let
commonConfig = ./common/acme/client;
dnsServerIP = nodes: nodes.dnsserver.networking.primaryIPAddress;
dnsScript = nodes: let
dnsAddress = dnsServerIP nodes;
in pkgs.writeShellScript "dns-hook.sh" ''
set -euo pipefail
echo '[INFO]' "[$2]" 'dns-hook.sh' $*
if [ "$1" = "present" ]; then
${pkgs.curl}/bin/curl --data '{"host": "'"$2"'", "value": "'"$3"'"}' http://${dnsAddress}:8055/set-txt
else
${pkgs.curl}/bin/curl --data '{"host": "'"$2"'"}' http://${dnsAddress}:8055/clear-txt
fi
'';
dnsConfig = nodes: {
dnsProvider = "exec";
dnsPropagationCheck = false;
environmentFile = pkgs.writeText "wildcard.env" ''
EXEC_PATH=${dnsScript nodes}
EXEC_POLLING_INTERVAL=1
EXEC_PROPAGATION_TIMEOUT=1
EXEC_SEQUENCE_INTERVAL=1
'';
};
documentRoot = pkgs.runCommand "docroot" {} ''
mkdir -p "$out"
echo hello world > "$out/index.html"
'';
vhostBase = {
forceSSL = true;
locations."/".root = documentRoot;
};
vhostBaseHttpd = {
forceSSL = true;
inherit documentRoot;
};
simpleConfig = {
security.acme = {
certs."http.example.test" = {
listenHTTP = ":80";
};
};
networking.firewall.allowedTCPPorts = [ 80 ];
};
# Base specialisation config for testing general ACME features
webserverBasicConfig = {
services.nginx.enable = true;
services.nginx.virtualHosts."a.example.test" = vhostBase // {
enableACME = true;
};
};
# Generate specialisations for testing a web server
mkServerConfigs = { server, group, vhostBaseData, extraConfig ? {} }: let
baseConfig = { nodes, config, specialConfig ? {} }: lib.mkMerge [
{
security.acme = {
defaults = (dnsConfig nodes);
# One manual wildcard cert
certs."example.test" = {
domain = "*.example.test";
};
};
users.users."${config.services."${server}".user}".extraGroups = ["acme"];
services."${server}" = {
enable = true;
virtualHosts = {
# Run-of-the-mill vhost using HTTP-01 validation
"${server}-http.example.test" = vhostBaseData // {
serverAliases = [ "${server}-http-alias.example.test" ];
enableACME = true;
};
# Another which inherits the DNS-01 config
"${server}-dns.example.test" = vhostBaseData // {
serverAliases = [ "${server}-dns-alias.example.test" ];
enableACME = true;
# Set acmeRoot to null instead of using the default of "/var/lib/acme/acme-challenge"
# webroot + dnsProvider are mutually exclusive.
acmeRoot = null;
};
# One using the wildcard certificate
"${server}-wildcard.example.test" = vhostBaseData // {
serverAliases = [ "${server}-wildcard-alias.example.test" ];
useACMEHost = "example.test";
};
};
};
# Used to determine if service reload was triggered
systemd.targets."test-renew-${server}" = {
wants = [ "acme-${server}-http.example.test.service" ];
after = [ "acme-${server}-http.example.test.service" "${server}-config-reload.service" ];
};
}
specialConfig
extraConfig
];
in {
"${server}".configuration = { nodes, config, ... }: baseConfig {
inherit nodes config;
};
# Test that server reloads when an alias is removed (and subsequently test removal works in acme)
"${server}-remove-alias".configuration = { nodes, config, ... }: baseConfig {
inherit nodes config;
specialConfig = {
# Remove an alias, but create a standalone vhost in its place for testing.
# This configuration results in certificate errors as useACMEHost does not imply
# append extraDomains, and thus we can validate the SAN is removed.
services."${server}" = {
virtualHosts."${server}-http.example.test".serverAliases = lib.mkForce [];
virtualHosts."${server}-http-alias.example.test" = vhostBaseData // {
useACMEHost = "${server}-http.example.test";
};
};
};
};
# Test that the server reloads when only the acme configuration is changed.
"${server}-change-acme-conf".configuration = { nodes, config, ... }: baseConfig {
inherit nodes config;
specialConfig = {
security.acme.certs."${server}-http.example.test" = {
keyType = "ec384";
# Also test that postRun is exec'd as root
postRun = "id | grep root";
};
};
};
};
in {
name = "acme";
meta = {
maintainers = lib.teams.acme.members;
# Hard timeout in seconds. Average run time is about 7 minutes.
timeout = 1800;
};
nodes = {
# The fake ACME server which will respond to client requests
acme = { nodes, ... }: {
imports = [ ./common/acme/server ];
networking.nameservers = lib.mkForce [ (dnsServerIP nodes) ];
};
# A fake DNS server which can be configured with records as desired
# Used to test DNS-01 challenge
dnsserver = { nodes, ... }: {
networking.firewall.allowedTCPPorts = [ 8055 53 ];
networking.firewall.allowedUDPPorts = [ 53 ];
systemd.services.pebble-challtestsrv = {
enable = true;
description = "Pebble ACME challenge test server";
wantedBy = [ "network.target" ];
serviceConfig = {
ExecStart = "${pkgs.pebble}/bin/pebble-challtestsrv -dns01 ':53' -defaultIPv6 '' -defaultIPv4 '${nodes.webserver.networking.primaryIPAddress}'";
# Required to bind on privileged ports.
AmbientCapabilities = [ "CAP_NET_BIND_SERVICE" ];
};
};
};
# A web server which will be the node requesting certs
webserver = { nodes, config, ... }: {
imports = [ commonConfig ];
networking.nameservers = lib.mkForce [ (dnsServerIP nodes) ];
networking.firewall.allowedTCPPorts = [ 80 443 ];
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
# OpenSSL will be used for more thorough certificate validation
environment.systemPackages = [ pkgs.openssl ];
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
# Set log level to info so that we can see when the service is reloaded
services.nginx.logError = "stderr info";
specialisation = {
# Tests HTTP-01 verification using Lego's built-in web server
http01lego.configuration = simpleConfig;
renew.configuration = lib.mkMerge [
simpleConfig
{
# Pebble provides 5 year long certs,
# needs to be higher than that to test renewal
security.acme.certs."http.example.test".validMinDays = 9999;
}
];
# Tests that account creds can be safely changed.
accountchange.configuration = lib.mkMerge [
simpleConfig
{
security.acme.certs."http.example.test".email = "admin@example.test";
}
];
# First derivation used to test general ACME features
general.configuration = { ... }: let
caDomain = nodes.acme.test-support.acme.caDomain;
email = config.security.acme.defaults.email;
# Exit 99 to make it easier to track if this is the reason a renew failed
accountCreateTester = ''
test -e accounts/${caDomain}/${email}/account.json || exit 99
'';
in lib.mkMerge [
webserverBasicConfig
{
# Used to test that account creation is collated into one service.
# These should not run until after acme-finished-a.example.test.target
systemd.services."b.example.test".preStart = accountCreateTester;
systemd.services."c.example.test".preStart = accountCreateTester;
services.nginx.virtualHosts."b.example.test" = vhostBase // {
enableACME = true;
};
services.nginx.virtualHosts."c.example.test" = vhostBase // {
enableACME = true;
};
}
];
# Test OCSP Stapling
ocsp-stapling.configuration = { ... }: lib.mkMerge [
webserverBasicConfig
{
security.acme.certs."a.example.test".ocspMustStaple = true;
services.nginx.virtualHosts."a.example.test" = {
extraConfig = ''
ssl_stapling on;
ssl_stapling_verify on;
'';
};
}
];
# Validate service relationships by adding a slow start service to nginx' wants.
# Reproducer for https://github.com/NixOS/nixpkgs/issues/81842
slow-startup.configuration = { ... }: lib.mkMerge [
webserverBasicConfig
{
systemd.services.my-slow-service = {
wantedBy = [ "multi-user.target" "nginx.service" ];
before = [ "nginx.service" ];
preStart = "sleep 5";
script = "${pkgs.python3}/bin/python -m http.server";
};
services.nginx.virtualHosts."slow.example.test" = {
forceSSL = true;
enableACME = true;
locations."/".proxyPass = "http://localhost:8000";
};
}
];
security/acme: limit concurrent certificate generations fixes #232505 Implements the new option `security.acme.maxConcurrentRenewals` to limit the number of certificate generation (or renewal) jobs that can run in parallel. This avoids overloading the system resources with many certificates or running into acme registry rate limits and network timeouts. Architecture considerations: - simplicity, lightweight: Concerns have been voiced about making this already rather complex module even more convoluted. Additionally, locking solutions shall not significantly increase performance and footprint of individual job runs. To accomodate these concerns, this solution is implemented purely in Nix, bash, and using the light-weight `flock` util. To reduce complexity, jobs are already assigned their lockfile slot at system build time instead of dynamic locking and retrying. This comes at the cost of not always maxing out the permitted concurrency at runtime. - no stale locks: Limiting concurrency via locking mechanism is usually approached with semaphores. Unfortunately, both SysV as well as POSIX-Semaphores are *not* released when the process currently locking them is SIGKILLed. This poses the danger of stale locks staying around and certificate renewal being blocked from running altogether. `flock` locks though are released when the process holding the file descriptor of the lock file is KILLed or terminated. - lockfile generation: Lock files could either be created at build time in the Nix store or at script runtime in a idempotent manner. While the latter would be simpler to achieve, we might exceed the number of permitted concurrent runs during a system switch: Already running jobs are still locked on the existing lock files, while jobs started after the system switch will acquire locks on freshly created files, not being blocked by the still running services. For this reason, locks are generated and managed at runtime in the shared state directory `/var/lib/locks/`. nixos/security/acme: move locks to /run also, move over permission and directory management to systemd-tmpfiles nixos/security/acme: fix some linter remarks in my code there are some remarks left for existing code, not touching that nixos/security/acme: redesign script locking flow - get rid of subshell - provide function for wrapping scripts in a locked environment nixos/acme: improve visibility of blocking on locks nixos/acme: add smoke test for concurrency limitation heavily inspired by m1cr0man nixos/acme: release notes entry on new concurrency limits nixos/acme: cleanup, clarifications
2023-07-18 10:20:33 +01:00
concurrency-limit.configuration = {pkgs, ...}: lib.mkMerge [
webserverBasicConfig {
security.acme.maxConcurrentRenewals = 1;
services.nginx.virtualHosts = {
"f.example.test" = vhostBase // {
enableACME = true;
};
"g.example.test" = vhostBase // {
enableACME = true;
};
"h.example.test" = vhostBase // {
enableACME = true;
};
};
systemd.services = {
# check for mutual exclusion of starting renew services
"acme-f.example.test".serviceConfig.ExecPreStart = "+" + (pkgs.writeShellScript "test-f" ''
test "$(systemctl is-active acme-{g,h}.example.test.service | grep activating | wc -l)" -le 0
'');
"acme-g.example.test".serviceConfig.ExecPreStart = "+" + (pkgs.writeShellScript "test-g" ''
test "$(systemctl is-active acme-{f,h}.example.test.service | grep activating | wc -l)" -le 0
'');
"acme-h.example.test".serviceConfig.ExecPreStart = "+" + (pkgs.writeShellScript "test-h" ''
test "$(systemctl is-active acme-{g,f}.example.test.service | grep activating | wc -l)" -le 0
'');
};
}
];
# Test lego internal server (listenHTTP option)
# Also tests useRoot option
lego-server.configuration = { ... }: {
security.acme.useRoot = true;
security.acme.certs."lego.example.test" = {
listenHTTP = ":80";
group = "nginx";
};
services.nginx.enable = true;
services.nginx.virtualHosts."lego.example.test" = {
useACMEHost = "lego.example.test";
onlySSL = true;
};
};
2022-12-18 00:31:14 +00:00
# Test compatibility with Caddy
# It only supports useACMEHost, hence not using mkServerConfigs
} // (let
baseCaddyConfig = { nodes, config, ... }: {
security.acme = {
defaults = (dnsConfig nodes);
# One manual wildcard cert
certs."example.test" = {
domain = "*.example.test";
};
};
users.users."${config.services.caddy.user}".extraGroups = ["acme"];
services.caddy = {
enable = true;
security/acme: limit concurrent certificate generations fixes #232505 Implements the new option `security.acme.maxConcurrentRenewals` to limit the number of certificate generation (or renewal) jobs that can run in parallel. This avoids overloading the system resources with many certificates or running into acme registry rate limits and network timeouts. Architecture considerations: - simplicity, lightweight: Concerns have been voiced about making this already rather complex module even more convoluted. Additionally, locking solutions shall not significantly increase performance and footprint of individual job runs. To accomodate these concerns, this solution is implemented purely in Nix, bash, and using the light-weight `flock` util. To reduce complexity, jobs are already assigned their lockfile slot at system build time instead of dynamic locking and retrying. This comes at the cost of not always maxing out the permitted concurrency at runtime. - no stale locks: Limiting concurrency via locking mechanism is usually approached with semaphores. Unfortunately, both SysV as well as POSIX-Semaphores are *not* released when the process currently locking them is SIGKILLed. This poses the danger of stale locks staying around and certificate renewal being blocked from running altogether. `flock` locks though are released when the process holding the file descriptor of the lock file is KILLed or terminated. - lockfile generation: Lock files could either be created at build time in the Nix store or at script runtime in a idempotent manner. While the latter would be simpler to achieve, we might exceed the number of permitted concurrent runs during a system switch: Already running jobs are still locked on the existing lock files, while jobs started after the system switch will acquire locks on freshly created files, not being blocked by the still running services. For this reason, locks are generated and managed at runtime in the shared state directory `/var/lib/locks/`. nixos/security/acme: move locks to /run also, move over permission and directory management to systemd-tmpfiles nixos/security/acme: fix some linter remarks in my code there are some remarks left for existing code, not touching that nixos/security/acme: redesign script locking flow - get rid of subshell - provide function for wrapping scripts in a locked environment nixos/acme: improve visibility of blocking on locks nixos/acme: add smoke test for concurrency limitation heavily inspired by m1cr0man nixos/acme: release notes entry on new concurrency limits nixos/acme: cleanup, clarifications
2023-07-18 10:20:33 +01:00
virtualHosts."a.example.test" = {
useACMEHost = "example.test";
extraConfig = ''
root * ${documentRoot}
'';
};
};
};
in {
caddy.configuration = baseCaddyConfig;
# Test that the server reloads when only the acme configuration is changed.
"caddy-change-acme-conf".configuration = { nodes, config, ... }: lib.mkMerge [
(baseCaddyConfig {
inherit nodes config;
})
{
security.acme.certs."example.test" = {
keyType = "ec384";
};
2021-12-04 19:01:18 +00:00
}
];
# Test compatibility with Nginx
}) // (mkServerConfigs {
server = "nginx";
group = "nginx";
vhostBaseData = vhostBase;
})
# Test compatibility with Apache HTTPD
// (mkServerConfigs {
server = "httpd";
group = "wwwrun";
vhostBaseData = vhostBaseHttpd;
extraConfig = {
services.httpd.adminAddr = config.security.acme.defaults.email;
};
});
};
# The client will be used to curl the webserver to validate configuration
client = { nodes, ... }: {
imports = [ commonConfig ];
networking.nameservers = lib.mkForce [ (dnsServerIP nodes) ];
# OpenSSL will be used for more thorough certificate validation
environment.systemPackages = [ pkgs.openssl ];
};
};
testScript = { nodes, ... }:
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
let
caDomain = nodes.acme.test-support.acme.caDomain;
newServerSystem = nodes.webserver.config.system.build.toplevel;
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
switchToNewServer = "${newServerSystem}/bin/switch-to-configuration test";
in
2019-10-27 12:53:55 +00:00
# Note, wait_for_unit does not work for oneshot services that do not have RemainAfterExit=true,
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
# this is because a oneshot goes from inactive => activating => inactive, and never
# reaches the active state. Targets do not have this issue.
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
''
import time
TOTAL_RETRIES = 20
class BackoffTracker(object):
delay = 1
increment = 1
def handle_fail(self, retries, message) -> int:
assert retries < TOTAL_RETRIES, message
print(f"Retrying in {self.delay}s, {retries + 1}/{TOTAL_RETRIES}")
time.sleep(self.delay)
# Only increment after the first try
if retries == 0:
self.delay += self.increment
self.increment *= 2
return retries + 1
backoff = BackoffTracker()
def switch_to(node, name):
# On first switch, this will create a symlink to the current system so that we can
# quickly switch between derivations
root_specs = "/tmp/specialisation"
node.execute(
f"test -e {root_specs}"
f" || ln -s $(readlink /run/current-system)/specialisation {root_specs}"
)
switcher_path = f"/run/current-system/specialisation/{name}/bin/switch-to-configuration"
rc, _ = node.execute(f"test -e '{switcher_path}'")
if rc > 0:
switcher_path = f"/tmp/specialisation/{name}/bin/switch-to-configuration"
node.succeed(
f"{switcher_path} test"
)
# Ensures the issuer of our cert matches the chain
# and matches the issuer we expect it to be.
# It's a good validation to ensure the cert.pem and fullchain.pem
2023-05-20 03:10:21 +01:00
# are not still selfsigned after verification
def check_issuer(node, cert_name, issuer):
for fname in ("cert.pem", "fullchain.pem"):
actual_issuer = node.succeed(
f"openssl x509 -noout -issuer -in /var/lib/acme/{cert_name}/{fname}"
).partition("=")[2]
print(f"{fname} issuer: {actual_issuer}")
assert issuer.lower() in actual_issuer.lower()
# Ensure cert comes before chain in fullchain.pem
def check_fullchain(node, cert_name):
subject_data = node.succeed(
f"openssl crl2pkcs7 -nocrl -certfile /var/lib/acme/{cert_name}/fullchain.pem"
" | openssl pkcs7 -print_certs -noout"
)
for line in subject_data.lower().split("\n"):
if "subject" in line:
2021-05-08 17:38:00 +01:00
print(f"First subject in fullchain.pem: {line}")
assert cert_name.lower() in line
return
assert False
def check_connection(node, domain, retries=0):
result = node.succeed(
"openssl s_client -brief -verify 2 -CAfile /tmp/ca.crt"
f" -servername {domain} -connect {domain}:443 < /dev/null 2>&1"
)
for line in result.lower().split("\n"):
if "verification" in line and "error" in line:
retries = backoff.handle_fail(retries, f"Failed to connect to https://{domain}")
return check_connection(node, domain, retries)
def check_connection_key_bits(node, domain, bits, retries=0):
result = node.succeed(
"openssl s_client -CAfile /tmp/ca.crt"
f" -servername {domain} -connect {domain}:443 < /dev/null"
" | openssl x509 -noout -text | grep -i Public-Key"
)
print("Key type:", result)
if bits not in result:
retries = backoff.handle_fail(retries, f"Did not find expected number of bits ({bits}) in key")
return check_connection_key_bits(node, domain, bits, retries)
def check_stapling(node, domain, retries=0):
# Pebble doesn't provide a full OCSP responder, so just check the URL
result = node.succeed(
"openssl s_client -CAfile /tmp/ca.crt"
f" -servername {domain} -connect {domain}:443 < /dev/null"
" | openssl x509 -noout -ocsp_uri"
)
print("OCSP Responder URL:", result)
if "${caDomain}:4002" not in result.lower():
retries = backoff.handle_fail(retries, "OCSP Stapling check failed")
return check_stapling(node, domain, retries)
def download_ca_certs(node, retries=0):
exit_code, _ = node.execute("curl https://${caDomain}:15000/roots/0 > /tmp/ca.crt")
exit_code_2, _ = node.execute(
"curl https://${caDomain}:15000/intermediate-keys/0 >> /tmp/ca.crt"
)
if exit_code + exit_code_2 > 0:
retries = backoff.handle_fail(retries, "Failed to connect to pebble to download root CA certs")
return download_ca_certs(node, retries)
start_all()
dnsserver.wait_for_unit("pebble-challtestsrv.service")
client.wait_for_unit("default.target")
client.succeed(
'curl --data \'{"host": "${caDomain}", "addresses": ["${nodes.acme.networking.primaryIPAddress}"]}\' http://${dnsServerIP nodes}:8055/add-a'
)
acme.systemctl("start network-online.target")
acme.wait_for_unit("network-online.target")
acme.wait_for_unit("pebble.service")
2019-10-27 12:53:55 +00:00
download_ca_certs(client)
2019-10-27 12:53:55 +00:00
# Perform http-01 w/ lego test first
with subtest("Can request certificate with Lego's built in web server"):
switch_to(webserver, "http01lego")
webserver.wait_for_unit("acme-finished-http.example.test.target")
check_fullchain(webserver, "http.example.test")
check_issuer(webserver, "http.example.test", "pebble")
# Perform renewal test
with subtest("Can renew certificates when they expire"):
hash = webserver.succeed("sha256sum /var/lib/acme/http.example.test/cert.pem")
switch_to(webserver, "renew")
webserver.wait_for_unit("acme-finished-http.example.test.target")
check_fullchain(webserver, "http.example.test")
check_issuer(webserver, "http.example.test", "pebble")
hash_after = webserver.succeed("sha256sum /var/lib/acme/http.example.test/cert.pem")
assert hash != hash_after
# Perform account change test
with subtest("Handles email change correctly"):
hash = webserver.succeed("sha256sum /var/lib/acme/http.example.test/cert.pem")
switch_to(webserver, "accountchange")
webserver.wait_for_unit("acme-finished-http.example.test.target")
check_fullchain(webserver, "http.example.test")
check_issuer(webserver, "http.example.test", "pebble")
hash_after = webserver.succeed("sha256sum /var/lib/acme/http.example.test/cert.pem")
# Has to do a full run to register account, which creates new certs.
assert hash != hash_after
# Perform general tests
switch_to(webserver, "general")
with subtest("Can request certificate with HTTP-01 challenge"):
webserver.wait_for_unit("acme-finished-a.example.test.target")
check_fullchain(webserver, "a.example.test")
check_issuer(webserver, "a.example.test", "pebble")
webserver.wait_for_unit("nginx.service")
check_connection(client, "a.example.test")
with subtest("Runs 1 cert for account creation before others"):
webserver.wait_for_unit("acme-finished-b.example.test.target")
webserver.wait_for_unit("acme-finished-c.example.test.target")
check_connection(client, "b.example.test")
check_connection(client, "c.example.test")
2019-10-27 12:53:55 +00:00
2021-03-15 19:25:49 +00:00
with subtest("Certificates and accounts have safe + valid permissions"):
# Nginx will set the group appropriately when enableACME is used
group = "nginx"
2021-03-15 19:25:49 +00:00
webserver.succeed(
f"test $(stat -L -c '%a %U %G' /var/lib/acme/a.example.test/*.pem | tee /dev/stderr | grep '640 acme {group}' | wc -l) -eq 5"
2021-03-15 19:25:49 +00:00
)
webserver.succeed(
f"test $(stat -L -c '%a %U %G' /var/lib/acme/.lego/a.example.test/**/a.example.test* | tee /dev/stderr | grep '600 acme {group}' | wc -l) -eq 4"
2021-03-15 19:25:49 +00:00
)
webserver.succeed(
f"test $(stat -L -c '%a %U %G' /var/lib/acme/a.example.test | tee /dev/stderr | grep '750 acme {group}' | wc -l) -eq 1"
2021-03-15 19:25:49 +00:00
)
webserver.succeed(
f"test $(find /var/lib/acme/accounts -type f -exec stat -L -c '%a %U %G' {{}} \\; | tee /dev/stderr | grep -v '600 acme {group}' | wc -l) -eq 0"
2021-03-15 19:25:49 +00:00
)
# Selfsigned certs tests happen late so we aren't fighting the system init triggering cert renewal
with subtest("Can generate valid selfsigned certs"):
webserver.succeed("systemctl clean acme-a.example.test.service --what=state")
webserver.succeed("systemctl start acme-selfsigned-a.example.test.service")
check_fullchain(webserver, "a.example.test")
check_issuer(webserver, "a.example.test", "minica")
# Check selfsigned permissions
webserver.succeed(
f"test $(stat -L -c '%a %U %G' /var/lib/acme/a.example.test/*.pem | tee /dev/stderr | grep '640 acme {group}' | wc -l) -eq 5"
)
# Will succeed if nginx can load the certs
webserver.succeed("systemctl start nginx-config-reload.service")
with subtest("Correctly implements OCSP stapling"):
switch_to(webserver, "ocsp-stapling")
webserver.wait_for_unit("acme-finished-a.example.test.target")
check_stapling(client, "a.example.test")
with subtest("Can request certificate with HTTP-01 using lego's internal web server"):
switch_to(webserver, "lego-server")
webserver.wait_for_unit("acme-finished-lego.example.test.target")
webserver.wait_for_unit("nginx.service")
webserver.succeed("echo HENLO && systemctl cat nginx.service")
webserver.succeed("test \"$(stat -c '%U' /var/lib/acme/* | uniq)\" = \"root\"")
check_connection(client, "a.example.test")
check_connection(client, "lego.example.test")
with subtest("Can request certificate with HTTP-01 when nginx startup is delayed"):
webserver.execute("systemctl stop nginx")
switch_to(webserver, "slow-startup")
webserver.wait_for_unit("acme-finished-slow.example.test.target")
check_issuer(webserver, "slow.example.test", "pebble")
2021-12-04 19:01:18 +00:00
webserver.wait_for_unit("nginx.service")
check_connection(client, "slow.example.test")
2021-12-04 19:01:18 +00:00
security/acme: limit concurrent certificate generations fixes #232505 Implements the new option `security.acme.maxConcurrentRenewals` to limit the number of certificate generation (or renewal) jobs that can run in parallel. This avoids overloading the system resources with many certificates or running into acme registry rate limits and network timeouts. Architecture considerations: - simplicity, lightweight: Concerns have been voiced about making this already rather complex module even more convoluted. Additionally, locking solutions shall not significantly increase performance and footprint of individual job runs. To accomodate these concerns, this solution is implemented purely in Nix, bash, and using the light-weight `flock` util. To reduce complexity, jobs are already assigned their lockfile slot at system build time instead of dynamic locking and retrying. This comes at the cost of not always maxing out the permitted concurrency at runtime. - no stale locks: Limiting concurrency via locking mechanism is usually approached with semaphores. Unfortunately, both SysV as well as POSIX-Semaphores are *not* released when the process currently locking them is SIGKILLed. This poses the danger of stale locks staying around and certificate renewal being blocked from running altogether. `flock` locks though are released when the process holding the file descriptor of the lock file is KILLed or terminated. - lockfile generation: Lock files could either be created at build time in the Nix store or at script runtime in a idempotent manner. While the latter would be simpler to achieve, we might exceed the number of permitted concurrent runs during a system switch: Already running jobs are still locked on the existing lock files, while jobs started after the system switch will acquire locks on freshly created files, not being blocked by the still running services. For this reason, locks are generated and managed at runtime in the shared state directory `/var/lib/locks/`. nixos/security/acme: move locks to /run also, move over permission and directory management to systemd-tmpfiles nixos/security/acme: fix some linter remarks in my code there are some remarks left for existing code, not touching that nixos/security/acme: redesign script locking flow - get rid of subshell - provide function for wrapping scripts in a locked environment nixos/acme: improve visibility of blocking on locks nixos/acme: add smoke test for concurrency limitation heavily inspired by m1cr0man nixos/acme: release notes entry on new concurrency limits nixos/acme: cleanup, clarifications
2023-07-18 10:20:33 +01:00
with subtest("Can limit concurrency of running renewals"):
switch_to(webserver, "concurrency-limit")
webserver.wait_for_unit("acme-finished-f.example.test.target")
webserver.wait_for_unit("acme-finished-g.example.test.target")
webserver.wait_for_unit("acme-finished-h.example.test.target")
check_connection(client, "f.example.test")
check_connection(client, "g.example.test")
check_connection(client, "h.example.test")
with subtest("Works with caddy"):
switch_to(webserver, "caddy")
webserver.wait_for_unit("acme-finished-example.test.target")
webserver.wait_for_unit("caddy.service")
# FIXME reloading caddy is not sufficient to load new certs.
# Restart it manually until this is fixed.
webserver.succeed("systemctl restart caddy.service")
check_connection(client, "a.example.test")
with subtest("security.acme changes reflect on caddy"):
switch_to(webserver, "caddy-change-acme-conf")
webserver.wait_for_unit("acme-finished-example.test.target")
webserver.wait_for_unit("caddy.service")
# FIXME reloading caddy is not sufficient to load new certs.
# Restart it manually until this is fixed.
webserver.succeed("systemctl restart caddy.service")
check_connection_key_bits(client, "a.example.test", "384")
domains = ["http", "dns", "wildcard"]
for server, logsrc in [
("nginx", "journalctl -n 30 -u nginx.service"),
("httpd", "tail -n 30 /var/log/httpd/*.log"),
]:
wait_for_server = lambda: webserver.wait_for_unit(f"{server}.service")
with subtest(f"Works with {server}"):
try:
switch_to(webserver, server)
# Skip wildcard domain for this check ([:-1])
for domain in domains[:-1]:
webserver.wait_for_unit(
f"acme-finished-{server}-{domain}.example.test.target"
)
except Exception as err:
_, output = webserver.execute(
f"{logsrc} && ls -al /var/lib/acme/acme-challenge"
)
print(output)
raise err
wait_for_server()
for domain in domains[:-1]:
check_issuer(webserver, f"{server}-{domain}.example.test", "pebble")
for domain in domains:
check_connection(client, f"{server}-{domain}.example.test")
check_connection(client, f"{server}-{domain}-alias.example.test")
test_domain = f"{server}-{domains[0]}.example.test"
with subtest(f"Can reload {server} when timer triggers renewal"):
# Switch to selfsigned first
webserver.succeed(f"systemctl clean acme-{test_domain}.service --what=state")
webserver.succeed(f"systemctl start acme-selfsigned-{test_domain}.service")
check_issuer(webserver, test_domain, "minica")
webserver.succeed(f"systemctl start {server}-config-reload.service")
webserver.succeed(f"systemctl start test-renew-{server}.target")
check_issuer(webserver, test_domain, "pebble")
check_connection(client, test_domain)
with subtest("Can remove an alias from a domain + cert is updated"):
test_alias = f"{server}-{domains[0]}-alias.example.test"
switch_to(webserver, f"{server}-remove-alias")
webserver.wait_for_unit(f"acme-finished-{test_domain}.target")
wait_for_server()
check_connection(client, test_domain)
rc, _s = client.execute(
f"openssl s_client -CAfile /tmp/ca.crt -connect {test_alias}:443"
" </dev/null 2>/dev/null | openssl x509 -noout -text"
f" | grep DNS: | grep {test_alias}"
)
assert rc > 0, "Removed extraDomainName was not removed from the cert"
with subtest("security.acme changes reflect on web server"):
# Switch back to normal server config first, reset everything.
switch_to(webserver, server)
wait_for_server()
switch_to(webserver, f"{server}-change-acme-conf")
webserver.wait_for_unit(f"acme-finished-{test_domain}.target")
wait_for_server()
check_connection_key_bits(client, test_domain, "384")
Fix letsencrypt (#60219) * nixos/acme: Fix ordering of cert requests When subsequent certificates would be added, they would not wake up nginx correctly due to target units only being triggered once. We now added more fine-grained systemd dependencies to make sure nginx always is aware of new certificates and doesn't restart too early resulting in a crash. Furthermore, the acme module has been refactored. Mostly to get rid of the deprecated PermissionStartOnly systemd options which were deprecated. Below is a summary of changes made. * Use SERVICE_RESULT to determine status This was added in systemd v232. we don't have to keep track of the EXITCODE ourselves anymore. * Add regression test for requesting mutliple domains * Deprecate 'directory' option We now use systemd's StateDirectory option to manage create and permissions of the acme state directory. * The webroot is created using a systemd.tmpfiles.rules rule instead of the preStart script. * Depend on certs directly By getting rid of the target units, we make sure ordering is correct in the case that you add new certs after already having deployed some. Reason it broke before: acme-certificates.target would be in active state, and if you then add a new cert, it would still be active and hence nginx would restart without even requesting a new cert. Not good! We make the dependencies more fine-grained now. this should fix that * Remove activationDelay option It complicated the code a lot, and is rather arbitrary. What if your activation script takes more than activationDelay seconds? Instead, one should use systemd dependencies to make sure some action happens before setting the certificate live. e.g. If you want to wait until your cert is published in DNS DANE / TLSA, you could create a unit that blocks until it appears in DNS: ``` RequiredBy=acme-${cert}.service After=acme-${cert}.service ExecStart=publish-wait-for-dns-script ```
2019-08-29 15:32:59 +01:00
'';
}